Table of Contents
Stirling Revisited: Practical Approaches to Merging Two Systems Thinking Streams
‘Susan Howick, ‘Fran Ackermann and “David Andersen
"Dept of Management Science
University of Strathclyde
Rockefeller College
University at Albany
Keywords: Group Model Building, Scenarios, Cause Maps, Causal Influence Diagrams, Group
Support Systems, Strategy Modeling
Abstract
The 1994 International System Dynamics Conference, held in Stirling, reviewed a range of
related Systems Thinking approaches. This paper focuses on the specific approach described by
Eden in Stirling and proposes a number of guidelines that can be used to explicitly and formally
link Eden's Systems Thinking approaches to formal simulation models. The specific case
presented involves linking semantically rich scenario maps to a formal causal influence diagram
that was in turn used as the basis for a formal simulation model. While the case reported on is
quite specific, we suggest that a broader range of complementary systems thinking approaches
can and should be integrated with more traditional SD simulation methods. The specific case
study reported on examines a scenario-based simulation of the promotion of renewable energy
sources in the UK electric power market. This work also informs on-going research in group
model building, strategy modeling (especially using scenarios) and the on-going debate about
qualitative vs. quantitative system dynamics.
Introduction
The last International System Dynamics conference to be held in the UK was at Stirling in 1994.
During this conference, the System Dynamics community explored a range of complementary
“systems thinking” and “soft systems” methodologies and their relationship to more traditional
system dynamics simulation models (Richardson 1994; Eden 1994). The discussions covered
issues such as how soft systems methodologies and system dynamics can inform one another,
how far qualitative systems thinking can take us without using quantitative system dynamics and
how complementary systems thinking approaches can support one another. One of the themes,
picked up by Eden (1994), was the integration of system dynamics with another systems thinking
approach. This paper builds on this work by looking at practical approaches to merging two
systems thinking streams when building a formal system dynamics model.
The need to merge two systems thinking streams arises when building a formal system dynamics
model because it is generally the case that a high degrees of collaboration is required between
two sets of ‘experts’ with very different domains of technical expertise. Typically, client groups
know their own substantive area and can tell the story of their system eloquently using ordinary
language. On the other hand a modeler or team of modelers typically possess the special skills
necessary to cast these system stories into a running model. Modelers’ specialized expertises rest
with the numerous technical decisions that “translate” system stories into model components such
as stocks and flows, parameters, functions and run specifications.
Group oriented approaches to model building and strategy development frequently place a diverse
client stakeholder group in the same room with modelers (often supported with online computing)
so that a rich dialogue can occur between these two domains of expertise.
Group model building (Vennix, 1996; Vennix et a/., 1997) thus relies on a diverse set of mapping
technologies and projected or drawn images that place easy-to-understand-and-modify images of
the system before the client group so that the group can discuss, modify, and elaborate them. The
maps often include stakeholder maps, causal influence diagrams, system flow diagrams, concept
models, issue maps, sketches of formal non-linear mathematical relationships, workbooks of
system structure, and sometimes even model equations. These various maps serve as “boundary
objects” (Black, 2002; Zagonel, 2002) that facilitate discussions between clients as well as
between the clients and the modeling team.
However, at some point in most, if not all, group model building efforts, modelers and clients
experience a “gap” in these facilitated discussions. Although the starting point for the modeling
exercise is through the use of maps that are rich in semantic content and easy for clients to
interact with directly, other boundary objects that are necessary for creating formal models can
often be seen as obscure or opaque. For example, most client groups find it intuitive and easy to
create clusters of actors that interact within the system when completing a stakeholder map (Finn,
1996; Bryson et al., 2001; Ackermann and Eden, 2003). However, another form of boundary
object - more formal models created following special syntactic requirements (such as making
formal distinctions between stocks and flows or carefully distinguishing between the polarity of
positive and negative linkages) - do not match the ordinary language of client groups.
Hence, group model building sessions must deal with this distinction between system maps with
high semantic content that are easy for clients to relate to versus system models with high
syntactic requirements that are relatively opaque. As most group model building projects start
with qualitative maps of high semantic content and end with elaborated system views that obey
all of the syntactic requirements of the simulation language modelers strive to “bring along” the
client group through these stages of increasing precision and structuring of system maps.
Unfortunately, most group modeling approaches eventually involve a “leap of faith” in this
complicated process of translating between rich qualitative maps and system maps.
This research is focused on exploring means of managing this problem of how to create a
sequence of boundary objects (maps of system issues, policies, and structures) and facilitation
guidelines that allow more of the transfer from qualitative maps to highly structured models to
occur out in the open and in view of client groups. We seek to gain greater client involvement in
these steps of model formulation and specification. Such a process helps to provide a promising
future for the integrated use of complementary systems thinking approaches as discussed in
Stirling in 1994. The paper will explore some of the conceptual bases underpinning this work
before discussing the methodology adopted, the case used to support our research and the insights
gained.
Three Threads of Research which Inform This Work
Three distinct research areas are drawn upon to consider and develop a more transparent
procedure. They are a) group model building, b) soft or qualitative systems, and c) strategy
development and simulation.
With the advent of icon-oriented software that can allow models to be built rapidly and projected
in front of client groups, (Vennix, 1996; Vennix et a/., 1997; Richardson and Andersen, 1995;
Andersen et al., 1997; Rouwette, 2003) group model building has emerged as an important way
to quickly elicit model structure from a client group. Andersen and Richardson (1997) have
suggested that group modelers can advance work in that field by creating “scripts” or detailed
descriptions of modeler and facilitator behavior while in front of groups to both codify and
exchange information on how to do this type of work. Our work continues in that line by
proposing guidelines for incorporating semantically rich descriptions of scenarios into a formal
simulation model.
A second recent trend in the field has been to more clearly distinguish between “systems
thinking” versus a more traditional view of system dynamics as supported by formal computer
simulation models.' This was also explored at the 1994 conference in Stirling. Many of the soft
systems thinking approaches share in common the creation of word-and-arrow maps that capture
the thinking of client groups. Most recently Wolstenholme (1999), Coyle (2000,2001), Homer
and Olivia (2001), and Richardson (2001) have raised the issue of when and how it is necessary
to move from qualitative to quantitative system dynamics, where quantitative system dynamics
are understood to involve formal computer simulation. This research may inform these
discussions in the literature because we propose and test formal methods for bridging the gap
between qualitative system maps and quantitative simulation models.”
Finally, Warren (2002), Morecroft (1984, 1985), Morecroft and Sterman (1994) and Milling
(2002) among others have demonstrated that formal system dynamics models can be powerful
supports for strategic planning. Scenario analysis (Van der Heijden, 1996; Eden and Ackermann,
1998; Wack, 1985; Schwartz, 1991) is a crucial tool for strategic planning often revealing counter
intuitive effects and dynamic behavior and this research demonstrates how formal scenario
analysis can be explicitly linked to system dynamics simulation models using group friendly
approaches.
The research discussed in this paper not only aims to integrate the three distinct research areas
detailed above, but also to inform work that is being carried out in each of them. Before
discussing the research methodology adopted, the next section aims to provide the reader with
background knowledge of the project that provided the opportunity for the research to be
undertaken.
The Modeling Context: Renewable Obligation Certificates (ROCs) in the UK Power
Market
In 2002, the UK Government brought in legislation designed to encourage the development of
additional renewable energy sources within the electric power market in the UK. The Renewable
Obligation Certificate (ROC) plan mandated renewal energy targets for all suppliers of electricity
in the UK starting at 3.0% of total generation capacity in 2002 and rising to 10.4% by 2010.
Suppliers could meet these targets either by submitting ROCs certifying that a generation source
met policy guidelines or by “buying-out” of their mandated obligations at a price of 30 pounds
per Mega-Watt Hour (increasing with inflation). Buy-Out fees would be returned to all holders of
' Senge (1990) popularized “systems thinking” a non-simulation brand of system dynamics as the “Fifth
Discipline” necessary to create learning organization:
Indeed, the strategy maps that we report on in this literature are based directly on the “systems thinking”
approach described by Eden at the Stirling conference.
valid ROCs in direct proportion to how many ROCs each supplier had submitted. The legislation
also allowed for ROCs to be traded on a free market (OFGEM, 2002; Smith and Watson, 2002).
During the winter of 2003 a consortium of Scottish investors, planners, and consultants
approached the University of Strathclyde to complete an analysis of future scenarios that could
impact on the success or failure of new renewal energy investments under the ROC plan. The
project was seen to be a pilot project, with its output being used to gather interest in a much fuller
study. The purpose of the project was to elicit a set of scenarios that might lead to unexpected
outcomes in the newly created ROC market. Such scenarios could range from unexpected boosts
to profit (perhaps triggered by a sudden rise in the price of fossil fuel) to possible collapses in the
market due to forces such as over supply, changes in regulation, or unforeseen health and safety
issues. The project envisioned that the dynamic implications of the major scenarios would be
implemented within a simulation model. A system dynamics model was one of the candidate
simulation approaches.
Research Methodology
The research team assembled at Strathclyde University brought a unique set of skills to this
project. One member of the modeling team was expert at strategy modeling, especially the
elicitation of scenarios using Group Explorer (Eden and Ackermann, 1999) as a group facilitation
software support (Ackermann and Eden, 2001). The other member of the team was expert in
traditional system dynamics simulation. While the members of the modeling team were relatively
more expert in one of the techniques and approaches being used, both members had considerable
experience with all of the modeling methods being used (through work on previous projects),
thereby facilitating discussions of how to merge methods in a group setting. Thus, both modelers
had an in-depth understanding of the theories supporting their particular area of expertise, an
appreciation of the other’s area and extensive practical experience of employing the modeling
techniques in practice. From this basis, the process of developing potential guidelines was able to
take place in an informed and thoughtful manner avoiding any potential conflicts in mixing
methodologies (Mingers and Gill, 1997).
A second important feature of this project is that another senior modeler was assigned a strict
participant-observer role for the duration of the project. His role was to serve as a process
observer during all of the group meetings, producing detailed notes concerning interactions in the
room during the group modeling sessions and determining whether the intended design matched
that that was followed during the workshop (Argyris, 1982; Argyris and Schon, 1974). Between
modeling sessions, this modeler-reflector participated in the design of the guidelines that were
ultimately used to link the scenario maps with the system dynamics simulation allowing the team
to tap into his knowledge of group model building. The research reported in this article results
from this assignment of an independent participant observer to record and reflect on the overall
process (Luna and Andersen, 2004).
The project in detail
The entire project took place within a number of overlapping phases of work. The project
commenced with rapid scenario and simulation model development, arriving at a final product
within four months. Periods of time were scheduled where direct contact with the client group
was necessary to enable group model building to take place. Other times were reserved for
interviews, phone contact, or off-line modeling and report generation.
The project as initially envisioned was budgeted for ~35 person-days. Time with the client group
(or subsets of the client group) was limited to four half-day meetings. This was to accommodate
the need of members of the group to travel on meeting days. Six or seven members of the client
team attended the first and second meetings, while all thirteen participants attended the third and
fourth meetings.
The project involved 3 major phases of work. The first phase, surfacing and structuring of
material, commenced with the elicitation of key events that would define possible future ROC
scenarios. The focus question was straightforward: “What future events could have an
important impact on the financial performance of ROCs”. The two sessions (involving subgroups
of the client group) used a Group Decision Support System called Group Explorer to rapidly
collect these events (Ackermann and Eden, 2001). Working with a skilled facilitator, each
session sorted approximately 100 events into major themes and sought important consequences
for these clusters of events — essentially building up a cause map illustrating possible scenarios
(see Eden and Ackermann, 1999;Eden and Ackermann 1998 for more details).
Phase two involved the integration of the material gathered in the first phase. Following a process
of integration and analysis of the cause maps (Eden and Ackermann, 1998) the next workshop
involved the client team first reviewing and confirming the integrated scenarios, and then taking
away a workbook of “homework” from the meeting and returning the workbooks for further
analysis by the modelling team. This work represented an application of approaches worked out
and routinely used by Vennix (1996), Richardson and Andersen (1995) and others in group model
building projects. The third and final phase then moved onto classic model building, mostly done
“in the back room” with frequent consultation with experts nominated by the group and the
scenarios. The final scenarios along with their impact on the possible future of the ROC market
were then presented back to the entire client group for reflection and review of next steps.
Artefacts created in the Modeling Process
Figure | illustrates the sequence of artefacts that were created during the 3 major phases of the
modeling intervention. Some artefacts were created directly by group process in group
meetings—these are underlined. The modeling team created other artefacts while working “in the
back room” and not in view of the client group—these are in standard text that is non
italic/underlined. Finally, some artefacts created during one of the meetings were subsequently
brought into subsequent meetings and were discussed and modified (but not initially created) by
the client group—these are in italic.
Figure 1: Sequence of Boundary Objects Created During the Project
The objects in Figure 1 are organized in rough temporal sequence working from left to right. The
timing of the various group meetings is indicated at the bottom of the figure. The arrows
connecting the various products indicate that the object at the tail of the arrow was created in time
before the object at the head. In general the object at the tail of the arrow was used as a basis to
create the object at the head of the arrow. The heavy lines without arrowheads indicate that the
two objects at each end of these lines were crafted with significant amounts of interaction.
Three sequences of interconnected activity are also illustrated in Figure 1. Along the top of the
diagram is a sequence of products mainly associated with creating scenarios for the future of
ROCs. This sequence of products begins with the “Raw Scenario” maps elicited in the first
meeting and ends with the final definition of scenarios presented at the last meeting. Along the
bottom of the diagram is a sequence of objects associated with the creation of a formal model.
This begins with prior theory (relating to power markets and commodity cycles) and ends with
the final running model. In-between are objects whose purpose was to “traverse” between the
scenario building effort and the formal modeling project.
The key story that we want to tell in the rest of this paper is how the research team was able to get
the client group from the upper left hand corner where the project started with a Group Explorer
elicitation of scenario events down to the lower right hand corner. At the end of this chain, an
elaborated causal influence diagram was used to create a formal simulation model. These
beginning, middle and end points for our story are boxes with bold lines. Below, we proceed to
describe in brief detail the work processes and group facilitation techniques and modeling
guidelines that we used along this critical pathway.
Phase 1: Elicitation of Material
The initial meetings used Group Explorer to elicit participants’ perceptions of key events that
would have an important future impact on the price of ROCs. The process commenced with each
of the client team anonymously and simultaneously entering into the system possible events that
could trigger alternative futures. This allowed a wide range of contributions to be gathered very
quickly and ensured all had an equal share of airtime — ensuring that the procedure felt just (Kim
and Mauborgne, 1995) and enabling a wide range of perspectives to be elicited. The next stage
was spent sorting these events into major themes — essentially adopting a form of crude content
analysis. The final, and longest stage saw participants working together to build the events into a
means ends network (using cause mapping) — illustrating how major clusters of activity linked to
one another and what were their important consequences.
Figure 2: Photograph of one of the groups working on generating scenario events
Phase II: Integration of Material
(i) Integrating the Scenario maps to create one single semantically rich qualitative model
The “raw” scenario maps elicited in the first two meetings needed to be merged. Not surprisingly
there were common themes across the two workshops, for example local governmental policies or
newly emerging policies of the EU but there were also differences due to the different
compositions of the subgroups. The common themes provided an obvious starting point for the
integration with the remaining material being woven into the resultant structure (informed
through the observations of both the modelers and the participant observer). This process
involved working off line, whereby the modeling team focused on identifying scenario events and
where these were identical in meaning merging them (Eden and Ackermann, 1998). Where there
was some dispute over the exact meaning, causal links were drawn between the two events
illustrating potential connections but not reducing the richness and variety — the group would
check these new links during the next workshop. Once complete, various analytical routines
could be executed to determine particular model structures, for example the emergence of new
dominant themes or feedback loops.
(ii) Detecting feedback loops to highlight potentially dynamic behavior
As noted above, using the built in analytic features of the software, the modeling team was able to
identify any feedback loops. The next stage was to check whether the feedback loops were
genuine i.e. not a result of incorrect linking of scenario events. Each loop was subsequently
checked to determine whether the chain of argument ‘made sense’ followed by checking any
doubtful links or loops with experts in the field. Those feedback loops that remained were then
examined in detail. One area of interest was determining whether they comprised events from
only one workshop or whether they encompassed contributions from both of the workshops. In
addition, did they appear to contain contributions from different perspectives e.g. finance,
generation, supply etc. Finally consideration was given to reflecting on how detailed or nested?
they were. Feedback loops, which although focusing on a particular theme or process have a
number of different paths contributing towards the dynamic behavior, may have more impact or
likelihood of occurring than those relatively sparsely linked structures. The loops thus identified
were candidates for inclusion in the formal quantitative model structure. Due to the limited time
of the pilot project, these candidates were not incorporated into the simulation model, but their
impact on the behavior of the ROC market was discussed with the client group.
(iii) Building the Preliminary Causal Influence Diagram
In working to create a system dynamics model to simulate the scenarios that were being
produced, the team decided to construct a causal influence diagram to capture the main structure
of the ROC market. The literature suggests that there are a number of ways in which to get a
preliminary causal model. For example, a concept model (Richardson and Andersen, 1995) could
be produced early in the process or published models on markets which are believed to have a
similar structure could be used or a structure could be developed from the ground-up by
interviewing players in the market (Ackermann ef a/., 1997). For this project, the modelers had
been present at and had participated in the conversations in meetings one and two. Based on
those discussions, the modeling team believed that published work on electric power markets and
general commodity markets (for example, Larsen et al., 2001; Sterman, 2000; Ford, 1997) shared
features in common with the ROC market and therefore provided a good source from which to
construct a preliminary causal influence diagram. The preliminary causal influence diagram was
then modified through one-to-one discussions with experts in the ROC market. The resulting
diagram obeyed all the syntactic requirements of a formal causal influence diagram—indeed early
on in the project this causal influence diagram was recast as a preliminary running simulation
model (Saeed, 1998a, 1998b).
(iv) Working with Reference Mode Sketches
Before the third meeting, as part of building the causal influence diagram the modeling team also
had focused in on the price of ROCs and the relative supply versus demand of renewable
generation capacity as key variables that would be central to any reference mode for a final
system dynamics model. To get some feel for the values of these variables, at the third meeting,
the client team drew the reference modes for these two key variables using standard group
modeling scripts (Andersen and Richardson, 1997; Saeed, 1998a, 1998b)
(v) Reviewing the scenarios
The third meeting commenced with a review of the merged scenarios. This was to ensure that the
links integrating material from both workshops were valid as well as to develop further
> Nested feedback loops occur when there are many different routes making up and consolidating a single
feedback loop
integrating links and material. One of the insights gained from the analysis stage had been that the
scenarios tended to be very focused around one specific theme rather than built up of a series of
interacting events — the notion of interacting improbabilities (van der Heijden, 1996). Effort was
therefore spent eliciting how the events captured impacted not just the theme they were currently
supporting but also others — thus building a set of scenarios that addressed a broader spectrum.
(vi) Quantifying the key relationships
The final substantive portion of the third meeting centred on quantification of key relationships
necessary to complete a formal simulation model. Prior to the third meeting, the modelling team
had worked extensively with the preliminary causal mechanism and had identified 3 key
relationships that would need to be quantified before a running model could be constructed. The
modelling team used small group techniques to elicit these relationships from the client team
(Ford and Sterman, 1998; Lee at al, 1998)
(vii) Gaining Group Feedback — filling in a workbook
As the participant group left the third meeting, each member of the group was given a
‘workbook’ comprising five tasks. The first task requested that members reviewed the scenarios
checking further to ensure that they were ‘correct’ in terms of the links (and therefore suggesting
deletions and additions if need be). The second task requested a review of the causal influence
diagram, aiming to validate its structure. The third task was to take each scenario and suggest 3-5
links where the scenario impacted upon the causal influence diagram (this process is described in
more detail in the next section). For task 4 members were then asked to explain the importance of
the links made in task 3. The final task was a review of the quantification processes carried out in
the workshop (the above step). This material was then used to refine both the scenarios and the
causal influence diagram.
(viii) Determining the impact of the scenarios on the Causal Influence Diagram.
Merging the refined scenario event maps with a preliminary causal influence diagram created the
central boundary object used to organize much of the conversation during the third meeting. The
details of how this step took place represent the key product of this research effort.
By this stage in the modeling process, a coherent set of scenario maps had been produced along
with a causal influence diagram that captured the main dynamics that both the modeling and
client team believed to exist in the ROC market. The modeling process then focused on how the
various scenarios that had been developed would impact upon the causal influence diagram and
hence a resulting system dynamics model.
For this project the modeling team were keen to develop a replicable process that would clearly
demonstrate how they had moved from the qualitative scenario maps to their quantitative
implications. The team developed a set of guidelines that captured the process of linking the
scenarios to the causal influence diagram. We used these guidelines in working to link the
scenario maps with the causal influence diagram.
The main focus of the guidelines was to categorize each of the scenario events, in each of the
scenario maps, in one of the following categories:
a) Explanatory: the event was seen to be detailed material, that comprises examples or
elaboration for its consequences in the scenario map
b) Input to causal influence diagram: an event that directly impacts one or more of the
causal influence diagram variables (often supported by explanatory material)
c) Output from causal influence diagram: an event that is triggered by one or more of the
causal influence diagram variables
d) Aggregate: a less elaborated form of the full causal influence diagram structure: in
this case one or more of the linked scenario events provide an aggregated form of a more
elaborated structure in the causal influence diagram.
e) Causal influence diagram variable addition: the event suggests a structural change to
the causal influence diagram extending and elaborating it.
Based on this categorization of the scenario events, the modeling team undertook the following
steps:
Step 1: Begin at the Bottom of a Scenario Chain*: In order to take a systematic approach to
the categorization of the events, commence with the event at the ‘bottom’ of a scenario map i.e.
identify the longest chain of argument and begin with the most subordinate scenario event.
Step 2: Code the Scenario Variable: Determine which of the above 5 categories is the most
appropriate for the chosen event. Then, if the event
a) Explanatory, move up the next step in the chain of causality until reaching an event that
has more than one event leading into it. Consider the next most elaborated chain
supporting this event so as to address the entire portfolio.
b) An input to the causal influence diagram, link the event into the appropriate causal
influence diagram variable
c) An output from the causal influence diagram, link the appropriate causal influence
diagram variable to the event
d) A less elaborated form of the causal influence diagram, then simply make a note of
this
e) A causal influence diagram variable addition, alter the causal influence diagram
structure appropriately and link the scenario event to the new causal influence diagram
variable
Step 3: Repeat the Process: Start with the next most subordinate event chain and repeat step 2.
Steps 2 and 3 should be repeated until all events in the scenario map have been categorized and
the appropriate links made between the scenario map and the causal influence diagram.
An example illustrating how the above guidelines were used in practice is detailed below.
Figure 3 shows an excerpt from one of the scenario maps produced for the ROC project. The
scenario map is taken from the artefact ‘Merged and Refined Scenario Maps’ shown in Figure 1,
which is the result of merging the material elicited during the initial two meetings. Overall, 8
scenario maps were produced during the project and Figure 3 below shows only a small section of
one of these maps.
Figure 3 includes some of the events that the client group raised as contributing to a ‘General
reluctance to invest in power in the UK’. When reviewing this figure, please note the following:
e The numbers at the beginning of each event are simply reference numbers.
a scenario chain is the chain of argumentation linking events together in a causal manner. Thus the bottom
of the chain is the most sub-ordinate event.
e An arrow from event A to event B should be read as ‘event A may lead to event B’
e Ifanegative sign appears at an arrow head, then this should be read as ‘event A may not
lead to event B’
e The unbroken arrows shown in figure 3 only represent a small number of the links in and
out of each event. The dotted arrows with attached numbers represent links to further
events that are not displayed in this small excerpt.
e The different fonts used for the events relate to different categories of events that were
used as part of the analysis of the scenario maps, but are of no real significance to the
focus of this paper
Figure 3: Scenario events leading to a ‘General reluctance to invest in power in the UK’.
The results of the categorisation process for each of the events in Figure 3 is included in
Appendix I.
For the pilot project, it was agreed that the focus of the simulation model would be on those
events that acted as inputs to the causal influence diagram. The categorization process highlighted
3 events that acted as inputs for the scenario excerpt shown in Figure 3. Therefore, this led to the
scenario events being linked to the causal influence diagram as shown in Figure 4 below. In this
figure, the bold arrows represent the links between the scenario map and the causal influence
diagram. The top 5 concepts (i.e. those above the dotted line) are variables in the causal influence
diagram, whereas the bottom 4 concepts (i.e. those below the dotted line) are events from the
scenario map.
Figure 4: Links between scenario events and the causal influence diagram.
Although the process of coding the moves from the scenario maps to the quantitative model
structure was mainly carried out “in the backroom”, not in view of the client group, this was
mostly due to time pressures. However, the modeling team did lead the client group step-by-step
through the process for one of the scenarios. Prior to linking the two models the client group had
spent time discussing and modifying both the scenario map and the causal influence diagram in
order to enhance their understanding and ownership of both. The facilitators thus were able to
take each scenario event in turn and explain how they would be categorized in terms of the
guidelines discussed above. When a category was chosen for a scenario event, the appropriate
link to the causal influence diagram was made. This process led the client group to become
engaged in the linking process and resulted in them confirming the linkages and making
suggestions for other links that could be made between the scenario maps and the causal
influence diagram. This process of refining the linked maps also continued after the group session
as participants took a workbook of printed maps away with them to review and send back more
detailed comments. In addition, during the group session, the modeling group observed the client
group using both the scenario event names and variable names from the causal influence diagram
relatively comfortably in their discussions. The group appeared to move between the two sets of
concepts seamlessly.
10
The process aided the client group to observe a clear path from the original information gathered
from them at group sessions to the output from the system dynamics model. This was possible via
the scenario maps (built-up in a group session using events suggested by the group) to the links
between these events and the causal influence diagram, to a quantification of that model and
finally the results from simulating the quantitative model.
Phase III: Modeling
(i) Running Model
The modeling team working “in the back room” rather than with the client group constructed the
final running simulation model. As shown in Figure 1, in constructing the final running
simulation model, the modeling team made use of the updated scenarios, the reference mode
exercise, the results of the key quantification exercises, the refined causal mechanism, and the
“pearls of wisdom®” generated by the client group. In addition, the collective workbooks
contained detailed comments, modifications, and insights on most all of the material generated in
the three group meetings. Because such an extensive amount of material had been collected from
the client group, constructing the running simulation model to demonstrate project feasibility was
not a conceptually complicated process.
The running model was a system dynamics model that simulated the impact of different possible
future scenarios. The output from the model included time plots of major variables that
demonstrated the future behavior of the ROC market based on each of the scenarios. The pilot
project had created a preliminary proprietary model that could be used by investors to investigate
risk in the ROC market.
(ii) Final Presentation and Report
The results of the project were fed back to the client group through both a final presentation and
report. Both of these included an explanation of the methodology adopted, the steps involved in
each of the 3 phases of the project and a discussion of the varying time plots of major variables
that were output from the final running model. As the project had been planned as a pilot study,
the final presentation also included a session at the end to discuss possible future research
avenues and these ideas were included at the end of the final report.
Discussion
The above describes the team’s first time using the process to link scenario maps to a simulation
model. Since completing the project, the team has spent some time reflecting on the overall
project resulting in suggestions to improve the process.
The team believes that the use of a basic concept model earlier on in the process may be
advantageous. Such a model could be introduced to the client group during the initial couple of
meetings. The possible benefits from this are that it would introduce the client group to the idea
of what a simulation model is earlier in the process. Such an introduction may increase perceived
* These were important considerations that the participants felt that the modelling team should take into
account and were elicited at the end of the fourth meeting.
11
confidence in the causal influence diagram as a bridge between the scenario discussions and a
final running simulation model.
When showing the client group the links between the scenario maps and the causal influence
diagram, the group appeared to move between the two sets of variable names relatively
seamlessly. One reason for this may be that they were presented with the quantitative model
structure in terms of language and notation with which they are already familiar. For example, if
the model structure had been presented as a stock/flow diagram, then the move may not have
been as smooth. The quantitative model structure was presented in form that was identical to the
structure of the scenario maps that they had generated themselves. This similar visual form
enhanced client ownership of the causal influence diagram, thereby making the transition
between the semantically rich scenario maps and the high syntactic requirements of the causal
influence diagrams much easier.
There are 3 threads of research for which the work discussed in this paper has implications:
Group model building
This work informs and builds upon the work in group model building in three specific ways.
First, we present documentation of a case study that spans and merges two different ways of
doing group model building work. We believe that significant progress can be made when
practitioners working within different traditions have a chance to share their work procedures in
detail. Most importantly, our work draws attention to the broad diversity of boundary objects
(Black, 2002 and Zagonel 2002) that can come into play in a single model building intervention.
Second our work adds to the literature on group model building by adding a number of scripts
(Andersen and Richardson, 1997) for working with an expanded set of boundary objects as
shown in Figure | that can enlarge the realm of current practice. Finally, the work presents a set
of coding procedures that can be used between group sessions to link formal causal influence
diagrams with semantically rich maps generated by client groups. We believe that this last
objective may be the most important contribution of this work and additional experimentation
along these lines will be necessary to fully refine the proposed coding rules.
Soft systems vs. quantitative simulation
The research presented here provides a concrete example of how and why it may not be necessary
to view soft systems vs. quantitative simulation as either-or approaches. This builds directly on
discussions at the 1994 Stirling conference. In the example that we discussed, we began with
qualitative mapping techniques and wound up with a formal simulation model. Using the scripts
and coding rules that we have described, we believe that it will become more and more common
(and easier) for modelers to move back and forth between these two complementary approaches
to system modeling. This extends the work done on mixing methodologies — typically spanning
the qualitative/quantitative divide (Mingers and Gill, 1997) and opens up the possibility of
addressing a wider range of problem domains. In addition, the work provides a concrete example
that may inform the debate on mapping versus simulating (Wolstenholme, 1999; Coyle, 2000 and
2001; Homer and Olivia, 2001; Richardson, 2001)
Strategy development and simulation
12
With respect to the linked fields of strategy development and simulation, our major contribution
has been to demonstrate techniques that explicitly link formal scenario mapping techniques into
quantitative simulation modeling. Often when working in strategy arenas, scenarios are
considered an appropriate and necessary part of any modeling team’s work. However little work
to date has been spent on quantifying scenarios elicited from groups. Hence the overall analysis
may miss observing potentially critical dynamic effects. For example through using formal
simulation modeling it will be possible to explore dynamic impacts of scenarios — understanding
their causes and examining a range of mitigating actions. Or, through appreciating the likelihood
of the company’s performance decreasing following implementation of a strategy before seeing
an increase — something that will help managers gain confidence in the strategy rather than
making a premature U-turn (see Pettigrew ef a/.,2003 for more detail). These approaches will
allow modelers to get the most out of both of these approaches in applied work.
Although this work builds upon the discussions during the 1994 Stirling conference, it should be
noted that the results described in this paper still represent work in progress. The intention of this
paper is to report on a process that has been developed and tested on one client group. This
reported work builds upon two of the authors work as part of a team involved in the modeling of
disruption and delay in complex projects as a part of litigation (Ackermann ef al., 1997; Eden et
al., 2000; Williams et a/., 2003) linked to a related but similar stream of research in group model
building (Richardson and Andersen, 1995; Andersen and Richardson, 1997). Such cross-
fertilization of approaches was generative for this work and, as noted at Stirling, important for the
field as a whole to make progress. We hope that future opportunities will arise where the process
described in this paper can be refined. Such refinements will allow modelers to provide their
client groups with clearer transitions between the development of qualitative models such as
scenario maps and quantitative simulation models.
The discussions held at Stirling in 1994 were important to the advancement of appreciating how
other systems thinking methodologies relate to the more traditional system dynamics simulation
models. This paper has described a practical approach that can be used to further advance the
work in this area. While the case reported on is quite specific, we believe that a broader range of
complementary systems thinking approaches can and should be integrated with more traditional
SD simulation methods. This area of work is an important area in the future of system dynamics
and we are delighted that it is still on the system dynamics community’s agenda a decade after
Stirling.
13
References
Ackermann F, Eden C. 2001. Contrasting Single User and Networked Group Decision Support
Systems for Strategy Making. Group Decision and Negotiation 10: 47-66.
Ackermann F, Eden C. 2003. Stakeholders Matter: Techniques for their identification and
management. Proceedings of the Academy of Management conference, Best Papers Series,
Seattle.
Ackermann F, Eden C, Williams T. 1997. Modeling for Litigation: Mixing Qualitative and
Quantitative Approaches. Interfaces 27: 48-65.
Andersen DF, Richardson GP. 1997. Scripts for Group Model Building. System Dynamics Review
13(2): 107-129.
Andersen DF, Richardson GP, Vennix JAM. 1997. Group Model Building: Adding More Science
to the Craft. System Dynamics Review 13(2): 187-201.
Argyris C. 1982. Reasoning, Learning and Action. San Francisco: Jossey Bass.
Argyris C, Schon DA. 1974. Theories in Practice. San Francisco: Jossey Bass.
Black LJ. 2002. Collaborating Across Boundaries: Theoretical, Empirical, and Simulated
Explorations. Unpublished Thesis (Ph. D.), Massachusetts Institute of Technology.
Bryson JM, Cunningham GL, Lokkesmoe, KJ. 2001. What To Do When Stakeholders Matter:
The Case of Problem Formulation for the African American Men Project of Hennepin County,
Minnesota. Public Administration Review 62: 568-584.
Coyle RG. 2000. Qualitative and Quantitative Modelling in System Dynamics: Some Research
Questions. System Dynamics Review 16(3): 225-144.
Coyle RG. 2001. Rejoinder to Homer and Oliva. System Dynamics Review 17(4): 357-363.
Eden C. 1994. Cognitive Mapping and Problem Structuring for System Dynamics Model
Building. System Dynamics Review 10(2/3): 257-276.
Eden C, Ackermann F. 1998. Analysing and Comparing Idiographic Causal Maps. In: Eden C,
Spender, JC. (Eds.) Managerial and Organizational Cognition, pp. 192-209. London: Sage
Eden C, Ackermann F. 1998. Making Strategy: the Journey of Strategic Management. London:
Sage.
Eden C, Ackermann F. 1999. The Role of GDSS in Scenario Development and Strategy Making.
In: 6" International SPIRE /5" International Workshop on Groupware proceedings, Cancun
Mexico, California: IEEE Computer Society Los Alamitos.
Eden C, Williams TM, Ackermann F, Howick S. 2000. On the nature of disruption and delay
(D&D) in major projects. Journal of the Operational Research Society 51: 291-300.
Finn C. 1996. Utilizing stakeholder strategies for positive collaboration. In: Huxham CE. (Ed.)
14
Creating Collaborative Advantage, pp. 152-164. London: Sage
Ford A. 1997. System Dynamics and the Electric Power Industry. System Dynamics Review
13(1): 57-85.
Ford DN, Sterman JD. 1998. Expert Knowledge Elicitation to Improve Formal and Mental
Models. System Dynamics Review 14(4): 309-340.
Homer JB, Oliva R. 2001. Maps and Models in System Dynamics: A Response to Coyle. System
Dynamics Review 17(4): 347-355.
Kim WC, Mauborgne RA. 1995. A Procedural Justice Model of Strategic Decision Making.
Organization Science 6: 44-61.
Larsen, ER, Lomi A, Dyner I. 2001. Learning from the future. Global Energy Business.
May/June, 23-26.
Lee TP, Andersen DF, Richardson GP, Rohrbaugh J, Zagonel, A. 1998. A Judgement Approach
to Estimating Parameters in Group Model-Building Sessions: A Case Study of Social Welfare
Reform in Dutchess County. Paper presented at the 16th International Conference of the System
Dynamics Society, Quebec '98, Quebec City, Canada.
Luna L, Andersen DL. 2004 Collecting and Analyzing Qualitative Data for System Dynamics:
Methods and Models. System Dynamics Review 19(4).
Milling PM. 2002. Understanding and Managing Innovation Processes. System Dynamics Review
18(1): 73-86.
Mingers J, Gill A. 1997. Multi-methodology: The Theory and Practice of Combining
Management Science Methodologies. Chichester: Wiley.
Morecroft JDW. 1984.Strategy Support Models. Strategic Management Journal 5(3): 215-229.
Morecroft JDW. 1985. The Feedback View of Business Policy and Strategy. System Dynamics
Review 1(1): 4-19.
Morecroft JDW, Sterman JD. (Eds.). 1994. Modeling for Learning Organizations. Portland, OR:
Productivity Press.
Office of Gas and Electricity Markets (OFGEM) (2002). The Renewables Obligation - OFGEM’s
Procedures. OFGEM Report February.
Pettigrew A, Whittington R, Melin L, Sanchez-Runde C, van der Bosch FAJ, Ruigrok W,
Numagami T. 2003. Jnnovative Forms of Organizing. London: Sage.
Richardson, GP. 1994. Introduction: Systems Thinkers, Systems Thinking. System Dynamics
Review 10(2/3): 95-99.
Richardson GP. 2001. Mapping versus Modeling: THE Answer to the Debate. Paper presented at
the 19th International Conference of the System Dynamics Society, Atlanta, Georgia.
15
Richardson GP, Andersen DF. 1995. Teamwork in Group Model Building. System Dynamics
Review 11(2): 113-137.
Rouwette E. 2003. Group model building as mutual persuasion. Nijmegen, The Netherlands.
Wolf Legal Publishers.
Saeed K. 1998a. Constructing Reference Mode. Paper presented at the 16th International
Conference of the System Dynamics Society Quebec '98, Quebec City, Canada.
Saeed K. 1998b. Defining a Problem or Constructing a Reference Mode. SSPS Working Paper:
Worcester Polytechnical Institute.
Schwartz P. 1991. The Art of the Long View. New York: Doubleday.
Senge PM. 1990. The Fifth Discipline: the Art and Practice of the Learning Organization. New
York: Doubleday/Currency.
Smith A, Watson J. 2002. The Renewables Obligations — Can It Deliver? Tyndall Centre Briefing
Note No.4, April.
Sterman JD. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World.
Boston: Irwin/McGraw-Hill.
van der Heijden K. 1996. Scenarios: the art of strategic conversation. Chichester: Wiley.
Vennix JAM. 1996. Group Model Building: Facilitating Team Learning Using System Dynamics.
Chichester: Wiley.
Vennix JAM, Richardson GP, Andersen DF. (Eds.). 1997. Group Model Building. System
Dynamics Review 13(2).
Wack P. 1985. Scenarios, Uncharted Waters Ahead. Harvard Business Review Sept-Oct: 73-
90.
Warren K. 2002. Competitive Strategy Dynamics. United Kingdom: Wiley.
Williams T, Ackermann F, Eden, C. 2003. Structuring a Delay and Disruption claim: an
application of cause-mapping and System Dynamics. European Journal of Operational Research
148: 192-204.
Wolstenholme EF. 1999. Qualitative vs Quantitative Modelling: The Evolving Balance. Journal
of Operational Research 50(4): 422-428.
Zagonel do Santos AA. 2002. Model Conceptualization in Group Model Building: A Review of
the Literature Exploring the Tension Between Representing Reality and Negotiating a Social
Order. Paper presented at the Proceedings of the 20th International Conference of the System
Dynamics Society, Palermo, Italy.
16
Appendix I - Categorization process for each of the events included in Figure 3
Scenario event Categorization Discussion
number
(see Figure 3)
37, 95, 117,58, | Explanatory All examples of a specific situation that may cause
52. event 115
115 Input to causal Link to causal influence diagram via ‘Availability
influence diagram via | of finance for projects’. However, also a link to this
258 from 258 (see later). Describes the same story,
therefore this direct link is not required.
89 Input to causal Link to causal influence diagram via ‘Delay in
influence diagram planning permission’
123 Input to causal Link to causal influence diagram via ‘Delay in
influence diagram planning permission’ (but this already linked
(implies 2 inputs, but | through 89 which is an example of 123, so do not
due to previous require a second link) and to ‘Planning success rate’
linkage, only 1
required)
278 Less elaborated form | Link from this to 258: High returns means more
of existing causal investment — causal influence diagram links already
influence diagram clarify this
structure
261 Less elaborated for of | Link from 261 to 258 — Too much capacity means
existing causal less investment — causal influence diagram links
influence diagram already clarify this
structure
258 Input to causal Link to causal influence diagram variable
influence diagram ‘Availability of finance for projects’
Table 1: Categorization of the scenario events displayed in Figure 3
A fuller explanation of how the categorization process was carried out for some of the scenario
events is now given:
Event 37-Change of Government: The main impact that a change in Government was
believed to have on the ROC market, was that there would be uncertainty on the new
Government’s policies and hence their attitude to the ROC market. It was therefore not
believed that this should link directly to the causal influence diagram, but instead
influences it via event 115. It was therefore categorized as an explanatory event to event
115:
Event 258-General reluctance to invest in power in UK: If there is a general reluctance
to invest in power in the UK, the direct impact of this was believed to be that there would
be less finance available for power projects. This directly influences one of the causal
17
influence diagram variables i.e. ‘availability of project finance’. Event 258 was therefore
concluded to be an input to the causal influence diagram.
Event 278-See renewables make profit: If renewable energy projects are making a
profit, then it is believed that this will encourage investors to invest in future renewable
energy. Although this is captured in the scenario map extract, this argument is also
captured within the existing structure of the causal influence diagram. It was therefore
concluded that link between event 278 and event 258 was a less elaborated form of
existing causal influence diagram structure.
Two categories that are not covered by the above example are an output from the causal influence
diagram and a causal influence diagram variable addition. However, these categories were used to
categorize events from other scenario maps — not all maps contributing to all of the categories.
For example, a sequence of events that were included in the scenarios was that if the renewable
energy targets were exceeded, then this might lead to ROC prices being driven down. However,
the existing causal influence diagram structure would capture this event through a variable named
‘market price of ROCs’ and so the reduction in ROC market prices would be an output from the
causal influence diagram.
An example of a causal influence diagram variable addition occurred through a scenario event
named ‘diminished public support for renewables’. The impact of positive or negative public
support had not been included in the preliminary causal influence diagram. Therefore, if the
scenarios impacted by this event were to be fully explored, then additional structure would need
to be added to the causal influence diagram. Due to the limited time available for the pilot project,
this amendment was not made to the causal influence diagram; instead a note was made of this
change for future reference when and if a more detailed second stage of the project was to be
carried out.
18
“Raw" Scenario}
|Maps from Mtgs|
1&2
Modelers.
Listen to
Elicitation
Modelers.
Review Literature on
Power Markets
and Commodity
Cycles
Phase I: Elicitation
of Material
“Pearls of
|Wisdom” from
Group
Sketches
Key
Relationships
Quantification]
Phase II: Integration
of Material
Joint Discussion
of the Model,
Scenarios, and
Final Report
Group
Workbook
Feedback
Reference Mode
Phase III: Modeling
Figure 1: Sequence of Boundary Objects Created During the Project
Arrow Key:
Solid Arrows:
Bold lines:
Dashed Arrows: Object at tail could be used to create object as head, but
Object at the tail used to create object at head
was not used to do so in the project discussed in the
paper
Two connected objects created with significant 19
interaction
Box Key:
Boxes/standard text: Artefacts created in the ‘backroom’
Boxes/italics: Artefacts created in the ‘backroom’ but discussed
and modified by client group
Boxes/underlined text: Artefacts created directly by group process
Boxes/Bold border: Focus of research
Hexagons: Processes that contribute to creation of artefacts
Figure 2: Photograph of one of the groups working on generating scenario events
20
Figure 3: Scenario events leading to a ‘General reluctance to invest in power in the UK’.
104104808 1042020
258 General are
r 123 planning
reluctance to invest
in power in UK ** wables [R 5!
HA mF
renewables [R 55 35]
Po FA.
261 Overcapacity esp \
NN
| 89 Regs and Legs for
if funding on a PPA | offshore wind not
| developed quickly
| enough
19359 277
SOK ‘h 115 increase
uncertainty re
eae policy [R 60 26] <__
Wit
aspen
t
/
/
58 Scottish Exec go
it alone
_~
~ 37 Change of
Government
117 change in
Ministers
95 Major event slows
development of
sector to await HSE
and environmental
52 Lack of cohesion a
between Scottish “even [R382
Exec and Westminster. 7
21
21
Figure 4: Links between scenario events and the causal influence
1027 RO electricity
capacity under
construction
_
1026 investment in
RO Electricity
<
1034 delay in
planning permission
etc
1046 availability of
project finance
diagram.
~~ 1044 planning
success rate
nN
Causal Influence Diagram
258 General
reluctance to invest
in power in UK
Fa
2
115 increase
uncertainty re
policy
89 Regulations and
Legislation for
offshore wind not =>
developed quickly
enough
22
Back to the Top
Scenario Map
123 planning regime
‘encourages
renewables
CC BY-NC-SA 4.0